On the specification of noise in two agent-based asset pricing models